https://www.kimi.com/blog/kimi-k2-6
Article
-
Moonshot AI releases Kimi K2.6, open-source coding-focused model.
-
Targets long-horizon agentic coding tasks and multi-step reasoning.
-
Priced at $0.95 input / $4 output, competitive with frontier models.
Discussion
-
simonw tested it via OpenRouter; it creatively wrapped SVG output in animated HTML.
-
Vibes-testing puts it near Claude Sonnet 4.6; one user says on-par or better than Opus.
-
China open-sourcing frontier-grade models while US labs close-source draws pointed commentary.
-
Benchmarks vs. real-world: some find K2.6 slightly weaker than K2.5 on puzzle/trick tasks.
Discuss on HN